Threading Dislocation Density Reduction in SiGe Strain Relaxed Buffer Layers through a Preexisting Dislocation Reservoir

Silicon germanium strain relaxed buffer layers (SRB) enable the integration of strained Si or strained Ge layers that can be used for improving CMOS devices with enhanced carrier mobilities, optoelectronics or emerging applications like qubit devices. Due to the lattice mismatch between Si and Ge, the formation of misfit dislocations during strain relaxation is inevitable. Misfit dislocations are accompanied by threading dislocations that penetrate the surface and possibly affect the device performance. A reduction of threading dislocation densities (TDD) with for example graded buffer layers [1] therefore has been subject of extensive research in the past and remains an active field today. In an earlier work we showed, that by using a backside deposition process [2],[3] - initially intended for controlling the wafer bowing - preexisting dislocations that are introduced at the edge of the wafer can fundamentally change the relaxation process of SiGe buffer layers.[4] Instead of rapid nucleation and multiplication of dislocations at single sites that lead to the formation of misfit dislocation bundles and pile-ups through blocking, the dislocations in the reservoir at the wafer edge can start gliding immediately upon reaching the Matthews and Blakeslee critical thickness [5] and form an evenly distributed dislocation network without pile-ups (Fig. 1a). With this method, the threading dislocation density of low strain layers (2% Ge) can be reduced by one order of magnitude. In this work, we analyze the influence of the backside approach on the crystal quality of strain relaxed buffer layers with higher Ge concentrations more relevant for industrial applications. SiGe buffer layers were grown using a high temperature (>820 °C), atmospheric CVD process using SiH 2 Cl 2 and GeCl 4 as precursor gases on commercial polished 300 mm Si wafers and substrates that received a backside deposition prior to the front side buffer growth. For strain relaxed buffers with constant composition the largest relative TDD reduction is observed at low Ge concentrations (one order of magnitude at 2% Ge). The TDD reduction becomes more and more ineffective towards higher Ge concentrations (Fig. 1b) which is attributed to increased spontaneous nucleation of misfit dislocation half-loops. At 10% Ge there is no significant improvement observed. However, when moving from constant composition buffers to graded buffers to 25% or 70% Ge, the TDD is still reduced by approximately 1 × 10 5 cm -2 to 1.5 × 10 5 cm -2 and 5 × 10 5 cm -2 , respectively (Fig. 1c). This is attributed to the removal of misfit dislocation bundles and pile-ups, that would block gliding dislocations from efficiently relaxing strain by trapping them and therefore increasing the need for nucleation of additional loops that increase the TDD. The absence of dislocation pile-ups also improves the surface quality (visual haze) of the SRB (Fig. 1d,e) as inhomogeneities in the dislocation network caused by misfit dislocation bundles that would lead to deep troughs in the crosshatch pattern are reduced. This is also reflected in the FWHM from rocking curves measured by HRXRD which is reduced by 60% for SRB with high Ge concentrations indicating a vastly improved crystal quality. In conclusion, this work highlights, that the drastically changed relaxation mechanism in the early stage of buffer growth by introducing a dislocation reservoir at the edge of the wafer has lasting consequences for the crystal quality of SRB even at high Ge concentrations. The prevention of dislocation bundle formation leads to a reduction of pile-up density, TDD, surface roughness and an overall improved crystal quality. The presented approach opens up the opportunity for a further reduction of the TDD that is necessary for future industrial applications of SiGe SRB. [1] E. A. Fitzgerald, Y.-H. Xie, M. L. Green, D. Brasen, A. R. Kortan, J. Michel, Y.-J. Mii, and B. E. Weir, Appl. Phys. Lett. 59, 811 (1991). [2] P. Storck and M. Vorderwestner, US2009236696, (2009). [3] G. Kozlowski, O. Fursenko, P. Zaumseil, T. Schroeder, M. Vorderwestner, and P. Storck, ECS Transactions 50, 613 (2013). [4] L. Becker, P. Storck, T. Schulz, M. H. Zoellner, L. Di Gaspare, F. Rovaris, A. Marzegalli, F. Montalenti, M. de Seta, G. Capellini, G. Schwalb, T. Schroeder, and M. Albrecht, J. Appl. Phys. 128, 215305 (2020). [5] J. W. Matthews and A. E. Blakeslee, Journal of Crystal Growth 29, 273 (1975). Figure 1